WordNet_Embeddings

This repository contains the code needed to train WordNet embeddings. In particular, most of the code is taken from: https://colab.research.google.com/github/hybridnlp/tutorial/blob/master/02_knowledge_graph_embeddings.ipynb

Dependencies

This project uses the conda environment. In the root folder you can find the .yml file for the configuration of the conda environment and also the .txt files for the pip environment.

Usage

Setup

# build conda env
conda env create -f env.yml 
conda activate wordnet_embeddings
pip install -r env.txt

# install scikit-kge
git clone https://github.com/hybridNLP2018/scikit-kge
cd scikit-kge
pip install nose
python setup.py sdist
pip install dist/scikit-kge-0.1.tar.gz
cd ../

# clone holographic-embeddings
git clone https://github.com/mnick/holographic-embeddings

Data Pre-processing

In order to make the dataset, type the following command that creates the dataset: ./holographic-embeddings/data/wn30.bin.

python HE_wordnet_preprocessing.py

Model Training

In order to train the model:

cd holographic-embeddings
python kg/run_hole.py --fin data/wn30.bin --fout wn30_holE_500_150_0.1_0.2.bin --ncomp 150 --test-all 100

Pretrained embeddings

In order to save the embeddings:

cd ../
python HE_wordnet_postprocessing.py

Then, the embeddings are saved in a .pickle file: ./holographic-embeddings/wn30_holE_500_150_0.1_0.2_embeddings.pickle

Licenze

MIT